[MIPS] FPU affinity for MT ASE.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

include/asm-mips/cpu-features.h

@@ -40,7 +40,7 @@
 #define cpu_has_sb1_cache	(cpu_data[0].options & MIPS_CPU_SB1_CACHE)
 #endif
 #ifndef cpu_has_fpu
-#define cpu_has_fpu		(cpu_data[0].options & MIPS_CPU_FPU)
+#define cpu_has_fpu		(current_cpu_data.options & MIPS_CPU_FPU)
 #endif
 #ifndef cpu_has_32fpr
 #define cpu_has_32fpr		(cpu_data[0].options & MIPS_CPU_32FPR)

include/asm-mips/fpu.h

@@ -21,6 +21,10 @@
 #include <asm/processor.h>
 #include <asm/current.h>
 
+#ifdef CONFIG_MIPS_MT_FPAFF
+#include <asm/mips_mt.h>
+#endif
+
 struct sigcontext;
 struct sigcontext32;
 

include/asm-mips/processor.h

@@ -134,6 +134,12 @@ struct thread_struct {
 
 	/* Saved fpu/fpu emulator stuff. */
 	union mips_fpu_union fpu;
+#ifdef CONFIG_MIPS_MT_FPAFF
+	/* Emulated instruction count */
+	unsigned long emulated_fp;
+	/* Saved per-thread scheduler affinity mask */
+	cpumask_t user_cpus_allowed;
+#endif /* CONFIG_MIPS_MT_FPAFF */
 
 	/* Saved state of the DSP ASE, if available. */
 	struct mips_dsp_state dsp;
@@ -159,6 +165,12 @@ struct thread_struct {
 #define MF_N32		MF_32BIT_ADDR
 #define MF_N64		0
 
+#ifdef CONFIG_MIPS_MT_FPAFF
+#define FPAFF_INIT 0, INIT_CPUMASK,
+#else
+#define FPAFF_INIT
+#endif /* CONFIG_MIPS_MT_FPAFF */
+
 #define INIT_THREAD  { \
         /* \
          * saved main processor registers \
@@ -173,6 +185,10 @@ struct thread_struct {
 	 * saved fpu/fpu emulator stuff \
 	 */ \
 	INIT_FPU, \
+	/* \
+	 * fpu affinity state (null if not FPAFF) \
+	 */ \
+	FPAFF_INIT \
 	/* \
 	 * saved dsp/dsp emulator stuff \
 	 */ \

include/asm-mips/system.h

@@ -155,6 +155,37 @@ extern asmlinkage void *resume(void *last, void *next, void *next_ti);
 
 struct task_struct;
 
+#ifdef CONFIG_MIPS_MT_FPAFF
+
+/*
+ * Handle the scheduler resume end of FPU affinity management.  We do this
+ * inline to try to keep the overhead down. If we have been forced to run on
+ * a "CPU" with an FPU because of a previous high level of FP computation,
+ * but did not actually use the FPU during the most recent time-slice (CU1
+ * isn't set), we undo the restriction on cpus_allowed.
+ *
+ * We're not calling set_cpus_allowed() here, because we have no need to
+ * force prompt migration - we're already switching the current CPU to a
+ * different thread.
+ */
+
+#define switch_to(prev,next,last)					\
+do {									\
+	if (cpu_has_fpu &&						\
+	    (prev->thread.mflags & MF_FPUBOUND) &&			\
+	     (!(KSTK_STATUS(prev) & ST0_CU1))) {			\
+		prev->thread.mflags &= ~MF_FPUBOUND;			\
+		prev->cpus_allowed = prev->thread.user_cpus_allowed;	\
+	}								\
+	if (cpu_has_dsp)						\
+		__save_dsp(prev);					\
+	next->thread.emulated_fp = 0;					\
+	(last) = resume(prev, next, next->thread_info);			\
+	if (cpu_has_dsp)						\
+		__restore_dsp(current);					\
+} while(0)
+
+#else
 #define switch_to(prev,next,last)					\
 do {									\
 	if (cpu_has_dsp)						\
@@ -163,6 +194,7 @@ do {									\
 	if (cpu_has_dsp)						\
 		__restore_dsp(current);					\
 } while(0)
+#endif
 
 /*
  * On SMP systems, when the scheduler does migration-cost autodetection,